Using label-free Raman microscopy to predict therapeutic resistance of TNBC cells

本申请是为了响应特别利益通知（NOSI）而提交的，该通知被确定为 NOT-CA-23-045。 细胞异质性已经变得至关重要，限制了治疗方法的有效性， 癌症治疗细胞异质性的表现可以在生物化学中观察到。 （分子组成和结构）、形态和机械水平，最终影响细胞 功能当考虑非侵入性单细胞方法时，评估的主要方法 群体生物化学异质性基于细胞的荧光激活细胞分选（FACS 表面标记。虽然非常快速，但该方法仅定量由选择的特定分子。 用户;但是如果标记不可靠呢？例如，研究表明，基于 在CD 44表面标志物的FACS上，仅提供了其中5%显示干细胞样 行为那么，脂质或核酸化学中所有其他潜在的分子变化， 在流式细胞仪实验中是看不见的，这可能是细胞中区分特征的信息？ 在这个补充申请中，我们建议开发一种无标记的分子成像方法， 表征细胞拉曼指纹并训练卷积神经网络（CNN）来预测 异源人群中的治疗耐药性。我们将使用高速，无标签的非线性 拉曼散射，其通过量化细胞的丰度来捕获细胞的整体分子指纹。 代谢物，脂质，蛋白质和核酸，并将这些数据与CNN训练和功能性 药物筛选基于我们的初步研究结果，CNN将通过培养 不同的乳腺癌亚群与增加浓度的阿霉素和测量 活细胞和死细胞的拉曼指纹（Aim 1）。在目标1中开发的CNN将测试其 在两个亚群的共培养物中预测亚群治疗敏感性的能力（目的 2）。该项目不仅提供了一个新的采取表型异质性癌细胞，但它也将 与下游功能或分子检测（如肿瘤启动或RNAseq）完全兼容 测量.

This application is being submitted in response to the Notice of Special Interest (NOSI) identified as NOT-CA-23-045. Cellular heterogeneity has become critically important, limiting the effectiveness of therapeutics in cancer treatment. The manifestations of cellular heterogeneity can be observed at the biochemical (molecular composition and structure), morphological, and mechanical level, ultimately affecting cell function. When considering non-invasive single-cell methods, the primary methods to assess population biochemical heterogeneity are based on fluorescence activated cell sorting (FACS) of cell surface markers. While extremely rapid, this method quantifies only specific molecules chosen by the user; but what if the marker is unreliable? For example, studies have shown that enriching cells based on FACS of CD44 surface markers only provides a population of cells in which 5% show stem-like behavior. What about all the other potential molecular changes in lipids or nucleic acid chemistry that were invisible in the FACS experiment, which might be informative as differentiating features in cells? In this supplement request, we propose to develop a label-free, molecular imaging approach to characterize cellular Raman fingerprints and train a convolutional neural network (CNN) to predict therapeutic resistance in heterogenous populations. We will use high-speed, label-free nonlinear Raman scattering, which captures a holistic molecular fingerprint of a cell by quantifying abundance of metabolites, lipids, proteins, and nucleic acids and couple this data with CNN training and functional drug screening. Building off our preliminary findings, a CNN will be developed and trained by culturing different breast cancer subpopulations with increasing concentrations of doxorubicin and measuring both live and dead cell Raman fingerprints (Aim 1). The CNN developed in Aim 1 will be tested for its ability to predict subpopulation therapeutic susceptibility in co-cultures of the two subpopulations (Aim 2). This project offers not only a new take on phenotyping heterogeneous cancer cells, but it will also be fully compatible with downstream functional or molecular testing such as tumor initiation or RNAseq measurements.